Breast Cancer Classification

Overview

This project applies basic machine learning methods to the Breast Cancer Wisconsin (Diagnostic) Dataset from the UCI repository.
The goal was to predict whether a tumor is malignant or benign based on cell nuclei measurements.

I used logistic regression with and without Principal Component Analysis (PCA) to evaluate how dimensionality reduction affects model performance.

---

Dataset

• Source: UCI Machine Learning Repository – Breast Cancer Wisconsin (Diagnostic)
• Size: 569 samples × 30 numeric features
• Target: Binary classification (Malignant = 1, Benign = 0)

---

Methods

1. Data Preprocessing
   • Standardized features (mean 0, unit variance)
   • Stratified train/test split
2. Models
   • Logistic Regression (baseline, no dimensionality reduction)
   • Logistic Regression after PCA (reduced to 2 components)
3. Evaluation
   • Accuracy on train and test sets
   • Confusion matrix and classification report
   • PCA scatter plot to visualize separation

---

Results

• Without PCA:
  • Train Accuracy ≈ 0.99
  • Test Accuracy ≈ 0.96
• With PCA (2 components):
  • Train Accuracy ≈ 0.96
  • Test Accuracy ≈ 0.95

PCA reduced dimensionality while retaining most variance.
Logistic regression still performed well, but raw features gave slightly higher accuracy.

---

Visualization

• PCA scatter plot shows malignant and benign samples forming distinct clusters.
• Confusion matrices highlight that both approaches achieve high precision/recall.

---

Takeaways

• Even simple models like Logistic Regression can achieve high accuracy on structured biomedical data.
• PCA is useful for visualization and dimensionality reduction, though it may slightly reduce accuracy.
• This project reflects how machine learning can support early cancer diagnosis, and demonstrates my ability to apply scikit-learn workflows to biomedical datasets.

---

Tech Stack

• Python
• scikit-learn
• NumPy, pandas, matplotlib

---

Repository

Full code and notebook: GitHub Repo